Proposal for next release: metadata/codes.parquet should be guaranteed to have all unique codes (or codes + code modifiers) for the dataset, even if other rows are empty.

[mmcdermott]

This makes it easier to possible to easily and safely rely on metadata/codes.parquet to perform inference over the codes included, such as to identify which codes match a regex filter, to use the set of codes as input to an LLM to automatically identify those associated with a certain concept, etc. Tagging @EthanSteinberg for your thoughts.

[EthanSteinberg]

I think this makes sense, as it's not too much of an annoyance and will make those operations much much easier